Undergraduate at MIT
20:30 UTC
Automatic differentiation (AD) is great: use gradients to optimize, sample faster, or just for fun! But what about coin flips? Agent-based models? Nope, these aren’t differentiable... or are they? StochasticAD.jl is an open-source research package for AD of stochastic programs, implementing AD algorithms for handling programs that can contain discrete randomness.